A Unified Uncertainty Representation for Graph Neural Networks via Doubly-Spectral Stochastic Expansion
One DSS representation is claimed to cover calibration, OOD scoring, and shift robustness for graph neural networks.
The paper models uncertain node embeddings as random graph signals, combining graph Fourier filters with an orthogonal-polynomial chaos coordinate. Its readouts use the mean coefficient for energy-based OOD scoring, higher-order coefficients for logit variation, and quadrature averaging for prediction and calibration. The authors report DSS-GNN leading Brier score results across 14 node classification benchmarks, while DSS-Hybrid performs strongest on most node-OOD settings and all 7 GOOD concept-shift benchmarks. The paper is listed as accepted at NeurIPS 2026. ArXiv · AI/CL/LG's note
The paper models uncertain node embeddings as random graph signals, combining graph Fourier filters with an orthogonal-polynomial chaos coordinate. Its readouts use the mean coefficient for energy-based OOD scoring, higher-order coefficients for logit variation, and quadrature averaging for prediction and calibration. The authors report DSS-GNN leading Brier score results across 14 node classification benchmarks, while DSS-Hybrid performs strongest on most node-OOD settings and all 7 GOOD concept-shift benchmarks. The paper is listed as accepted at NeurIPS 2026. ArXiv · AI/CL/LG's note
score 4